Home > Linux > Linux Kernel 2.4.26 Released Linux Kernel 2.4.26 Released Submitted by jeff 2004-04-14 Linux 47 Comments Linux kernel 2.4.26 was released today. You download it from kernel.org or from one of the mirrors. About The Author Adam Scheinberg Vice President, Information Technology at Massey Services, Inc • President, Board Member, The Mockingbird Foundation • All Things Web, Umphrey’s McGee • Web Developer • Father • Foodie • Music Snob • OS enthusiast Follow me on Twitter @sethadam1 47 Comments 2004-04-14 4:46 pm Who is actually still using 2.4? 2004-04-14 4:48 pm I do. Such as most people who feel that 2.6 is not really stable yet or those who want to spare themselves the hazzle of manually updating a kernel (I use Slackware, so please spare me of “but XYZ offers packages you §%$”). Face it: 2.4.x is still the standard kernel for most users at the moment as far as I see it. Hell, some still use 2.2 or 2.0, your question would have more merit there. 2004-04-14 4:56 pm > those who want to spare themselves the hazzle of manually updating a kernel > Face it: 2.4.x is still the standard kernel for most users at the moment Good, but who is going to use 2.4.26 anyway? As you said, 2.4 is the default kernel series in most distributions, but all distros that target at people who just want things to work (RedHat, SuSE, Mandrake) are shipping 2.6 now. Embedded systems (e.g. Zaurus) are still using older 2.4 kernels (OpenZaurus 3.3.5: 2.4.18-rmk7-pxa3-embedix), so my question remains: Who is going to use this kernel? 2004-04-14 4:57 pm the kernel 2.6 is to slow for that machine so i use 2.4 iam just in this time compiling 2.4.26 on the machine 2004-04-14 5:01 pm > the kernel 2.6 is to slow for that machine I thought 2.6 was supposed to be faster than 2.4… Have you tried 2.6? 2004-04-14 5:02 pm Who is actually still using 2.4? How about every sysadmin who runs Linux on a server that works? I don’t need any new features on our web servers, so why bother with a risky upgrade? I keep patched, run the latest 2.4 kernel, and I’m good to go. 2004-04-14 5:03 pm Good, but who is going to use 2.4.26 anyway? As you said, 2.4 is the default kernel series in most distributions, but all distros that target at people who just want things to work (RedHat, SuSE, Mandrake) are shipping 2.6 now. Embedded systems (e.g. Zaurus) are still using older 2.4 kernels (OpenZaurus 3.3.5: 2.4.18-rmk7-pxa3-embedix), so my question remains: Who is going to use this kernel? Xandros, Lindows, etc, those distros that “just work” all ship with 2.4-kernel. Fedora Core 1 has the 2.4-kernel afaik. Dozens of smaller distros still have the 2.4-kernel. Since small number updates are bugfixes why should those distros not put out that newer kernel as update while working on the next version with 2.6-kernel? Support for the 2.4-kernel does not just die instantly. 2004-04-14 5:07 pm I’m with Adam. I run a 2.4 kernel on my laptop. I tried 2.6, but there were a couple of minor hiccups, so I just went right back to 2.4, where the only things that are being released fix security, stability, and driver issues. Those are useful to me. My machine works great. Until I upgrade the whole distro, I see no reason to fiddle with another kernel. 2004-04-14 5:08 pm >I thought 2.6 was supposed to be faster than 2.4… Have you tried 2.6? Nope 2.6 is definitely slower. It was miserable on my pentium 200 mhz laptop. It may “feel” faster with gui applications on a fast box. But as far as I can tell it’s actually a bit slower overall. 2004-04-14 5:11 pm Nope 2.6 is definitely slower. It was miserable on my pentium 200 mhz laptop. It may “feel” faster with gui applications on a fast box. But as far as I can tell it’s actually a bit slower overall. That’s very interesting, have you tried doing any benchmarks and submitting results to the kernel team I’m sure they’d appreciate your input. I found the opposite 2.6 with the pre-emptive patch makes most areas of my system much more responsive. Doesn’t matter whether I’m at the command line or in X. Maybe this has something to do with pre 686 machines. 2004-04-14 5:26 pm >Who is actually still using 2.4? My faviorite live-cd, slax. because the overlay filesystem dosn’t support 2.6 yet. 2.6 is most important for large servers 2.4 is still faster for desktop use. plus, I don’t trust 2.6 to be stable or secure yet. there’s just been too many changes to it. 2.4 is still relevant and will still be for awhile. 2004-04-14 5:32 pm Gabriel, Just a few, very few people is using 2.6… everybody is using 2.4 and it will be used for a long time yet. You are assuming that every ws/server will be immediately updated just for the fact that new distros have been released. This is false. Nobody update the distro immediately, if you look carefully, the Enterprise version of RedHat Linux will ship 2.4 in the next major version. So… -William 2004-04-14 5:34 pm > Nope 2.6 is definitely slower. It was miserable on my pentium 200 mhz laptop. I have used same quite late 2.5 on 200 Mhz pentium, and it feeled faster than 2.4. But in fact, 2.6 default configuration is a bit slower than 2.4. For example, its probably better to set the timer to the old 100 Hz instead of 1000 Hz on slower machines. Also preemptible kernel is actually a bit slower (this can be disabled). 2004-04-14 5:36 pm It must be in use. Otherwise, why would they bother to update it? Heh. 2004-04-14 5:43 pm 2.4 still rocks hard. 2.6 is “stable” for home usage, but it just still changes too much from .minor version to another – it is not really stabilized yet to the stage that 2.4 has after three years of patching. 2004-04-14 5:48 pm [/i]Also preemptible kernel is actually a bit slower (this can be disabled).[/i] True. The mm-patchset have a low-latency patch that makes kernel preemption quite useless. 2004-04-14 5:51 pm No 2.6.* version has worked with my firewire drive. And all my posts to the linux-kernel mailing list about it have gone unanswered… Till they fix this problem, I have no intention of upgrading to 2.6, Thankfully, it works with 2.4.* and with FreeBSD. Adam 2004-04-14 6:01 pm I tried to install 2.6 but I would always screw up during “make menuconfig” I wish there was something like “make oldconfig” when upgrading from 2.4 to 2.6 2004-04-14 6:08 pm Just copy the .config file from your 2.4 directory to your 2.6 one. 2004-04-14 6:13 pm Does that really work? Have you tried it? 2004-04-14 6:18 pm well i have tried kernel 2.6 and it was slow on my machine so i use 2.4 now 2004-04-14 6:34 pm I remember asking that on http://www.justlinux.com and the answer was a decisive ‘No’. I likewise doubt it. 2004-04-14 6:35 pm 2.6 is slow and still has problems with X at least on the Fedora distro. I upgraded to 2.6.5 to give it another shot and now I get a kernel panic because I use Reiser. Besides, a lot of vendors like Netraverse don’t have 2.6 enabled kernels yet, I don’t believe. Too much of a pain. I’ll wait for Fedora Core 2 to switch to 2.6.x 2004-04-14 6:48 pm People who are still using 2.4 might be those who feel the 2.6 series is unstable, particularly the way they are adding new things like udev into a production kernel. They were doing this crap in 2.4 as well. Goes to show you that Linux is severely immature, and that BSD is the way to go if you want stable reliability that WORKS when it is released. 2004-04-14 7:02 pm Does anyone notice higher fps using 2.6 compared to 2.4? 2004-04-14 7:14 pm Hi “People who are still using 2.4 might be those who feel the 2.6 series is unstable, particularly the way they are adding new things like udev into a production kernel. They were doing this crap in 2.4 as well.” Udev was there in 2.5 seris and is being updated in the stable series. its not a new feature that was added in 2.6. dont spread misinformation 2004-04-14 7:21 pm I agree with you in the first paragraph… I use 2.6 (mdk10) and it is faster, I have no problems with it. But it is clear that it needs more real testing. Just remember that right now 2.6 is a lot much better than 2.4 was in the same stage of development. So if 2.4 rocks, 2.6 will rock even harder in the near future. -William 2004-04-14 7:40 pm so I’m still doing 2.4. Specifically, memory leaks for an rtl8150 device. Dies in about 15 minutes on its own, or rapidly with second USB device inserted. So it goes. 2004-04-14 8:14 pm I played ut2004 with kernel 2.4.22 (slackware 9.1) and it was a bit slower, then i have recently upgraded to 2.6.x and I get 10/15 more fps in that game, it’s much faster. sorry about my english… -linkcr 2004-04-14 8:18 pm Slackware 9.1/Kernel 2.6.5 on this AMD 32 box, no problems so far, snappy desktop performance. The pIII box downstairs does now have a few OpenGL glitches all of a sudden since upgrading that from 2.6.2 to 2.6.5 though. If you have a system that uses distro – patched kernels, only use kernels for your distro! (Fedora Core would be an example) 2004-04-14 8:28 pm If you have a system that uses distro – patched kernels, only use kernels for your distro! And people say that the BSDs are a fragmented forking party. Just how many of these distributions run differently patched kernels? As to too many things changing from one release to the next making the kernel unstable, that’s just a wonderful argument for microkernels – adding new functionality doesn’t mess with preexisiting kernel code, meaning that less testing is required to get from NTR to production ready. Too bad the Linux and the BSD folks just aren’t getting that simple concept. 2004-04-14 9:04 pm Hi “And people say that the BSDs are a fragmented forking party. Just how many of these distributions run differently patched kernels?” Many of the commerical distributions include patched kernels and backported features. a few patches arent called a fork ” As to too many things changing from one release to the next making the kernel unstable, that’s just a wonderful argument for microkernels – adding new functionality doesn’t mess with preexisiting kernel code, meaning that less testing is required to get from NTR to production ready. Too bad the Linux and the BSD folks just aren’t getting that simple concept. ” Microkernels look very good in paper. Real time implementations usually fail miserably. Can you name a few Mainstream operating systems using a good microkernel design? message based inter communication have severe performance disadvantages. layering is not always possible between architecture and clean seperation leads to poor interfaces for functionaly different entities 2004-04-14 9:11 pm why isn’t it working with your firewire drive? Do you have a cdrom or harddisk? In 2.6 my firewire cdrw and dvd player work fine. Do you understand that firewire works slightly differently in 2.6? I’m still running a 2.6-test kernel, but on my system, my firewire drive device files are located in /sys/bus/ieee1394/devices, could this change of device location be your problem? You should email me about it if you think I could be of any assistance. 2004-04-14 9:19 pm why isn’t it working with your firewire drive? I don’t know. I load the ieee1394 module, the ohci1394 module, and the sbp2 module, just as I’m supposed to. After/While it loads the sbp2 module, it sees the drive, starts spewing scsi errors, and is unable to read the partition table. It’ll literally be two or three minutes before the the “modprobe sbp2” command returns. “fdisk -l” then doesn’t show the firewire drive. Adam 2004-04-14 9:36 pm Actually, probably the best approach is to use a hybrid kernel, – essentially a microkernel that keeps some extra code in kernelspace when performance is absolutely critical. The stability and ease of adding features to a microkernel based OS on modern hardware more than offset the minor performance hit taken by the use of a message passing system. Most modern OSs do this; Linux and most of the BSDs and Solaris being the exceptions. But I guess it comes down to one thing; does the developer want to implement new features quickly and test them for ages just to keep the small performance advantage that monolithics have over microkernels, or do they want to implement new features quickly and not have to test quite so extensively due to the clean seperation of a microkernel design, while sacrificing just a little speed for a better system overall. Mach was a mess, and its popularity way back when is probably the biggest reason that people (Linux folks especially) are dead set against microkernel based OSs. Sure, there is a little overhead in message passing, but the benefits clearly outweigh that disadvantage regardless of whether you’re a developer or an end user. 2004-04-14 10:18 pm Hi Linux is a modular kernel like freebsd’s and it sure is a hybrid of the traditional monolothic or the fancy microkernel stuff. Message passing adds a HUGE amount of overheard when there are functional different systems like competing filesystem implementations in linux. imagine having a single interface for something as radical as reiserfs along with ext3 2004-04-14 11:22 pm Hell, we’re still using RedHat Linux 7.x. With 38 servers in 38 remote sites to look after, we don’t have the luxury of updating to every new piece of whatever that gets released. We pick and choose our software, stay up-to-date with security fixes, and leave things alone until the summer when school’s out (these servers are powering LTSP in elementary school labs). Most of our servers are running 2.4.20, but they’ll be updated to 2.4.26 over the next week or so to get better fs, ethernet, and sound support. We have no plans of moving to 2.6.x for another year or so, as all of our testing has shown it to be slower on our servers (dual-P3 1 GHz and dual-AthlonMP 2200+) than 2.4.x. When we do our major upgrade to either Debian or Fedora Core, then we’ll look at migrating over to 2.6.x. Or, maybe we’ll move everything over to FreeBSD and not worry about the upgrade-of-the-day syndrome. Not everyone runs a home computer. Not everyone needs the latest version of everything. And not everyone upgrades just for the sake of upgrading. Right now, 2.6 doesn’t offer us anything that 2.4 can’t. 2004-04-15 1:25 am People who are still using 2.4 might be those who feel the 2.6 series is unstable, particularly the way they are adding new things like udev into a production kernel. They were doing this crap in 2.4 as well. Heh… Not only udev isn’t in the kernel but it has nothing to do with it! It’s an userland application. Yes, 100% userland. Get at least some knowledge on the subject before spreading FUD. 2004-04-15 1:28 am That the same people that complain that Microsoft makes it’s users test thier software instead of the original developers, yet when ever someone in the “linux community” tells someone else about how slow 2.6 is the first thing the person answers: “That’s very interesting, have you tried doing any benchmarks and submitting results to the kernel team I’m sure they’d appreciate your input.” *hypocrisy* 2004-04-15 1:35 am Hi Give me the source of windows and you will get exactly the same treatment. 2004-04-15 2:45 am Linux is a modular kernel like freebsd’s and it sure is a hybrid of the traditional monolothic or the fancy microkernel stuff. No. Saying that it is a “hybrid” of the two kernel technologies implies that it shares mechanisms with both monolithic and microkernels. Modular monolithic kernels share nothing architechturaly with microkernels, and thus cannot be called a “hybrid” of the two. Message passing adds a HUGE amount of overheard Wrong again. There is a measurable, but insignificant overhead. Take DragonFly for example. More and more it is using a threaded message passing system, and there is overhead. But most of the overhead is from the serializing tokens as opposed to the message passing. It’s the same type of overhead you get by using mutexes in Linux or FreeBSD, and it operates independantly of the messaging system. when there are functional different systems like competing filesystem implementations in linux. I think I missed the turn off into crazyland somewhere… imagine having a single interface for something as radical as reiserfs along with ext3 Gawd knows they have absolutely nothing in common… 2004-04-15 3:42 am Well… I am still using 2.4 in some of the production servers, as well as machine that is using ATARAID. For those of us using ATARAID, you will notice this is not yet in 2.6 (officially)… But, I guess no many people use ATARAID right? Coz’ this seems to be the least complain thingy in 2.6… 2004-04-15 4:32 am You really do seem to like microkernels, but the only modern general purpose operating system in wide use using a microkernel are I guess Windows NT (look at where the “ease” of adding features has gotten them), and OSX. Versus Free/Net/OpenBSD, BSD OS, Linux, Solaris, AIX and most other proprietry UNIXes. Either way, I subscribe to the idea that *interfaces* are the most important thing to good modular and maintainable operating systems (and any sort of code)… not the way data is passed. With microkernels, one benefit can be that you have lots of code in seperate protected memory areas. In practice, performance constraints means that everyone has all their code moved into kernel mode anyway. So the difference becomes the underlying way that requests are made and data is moved around… you don’t see this anyway if you have good interfaces built on that. However in the microkernel case you still have the (however minimal) overhead of message passing. 2004-04-15 6:06 am This test is encouranging 2.6 series: http://www-106.ibm.com/developerworks/linux/library/l-web26/ 2004-04-15 12:48 pm Can you tell me where can I get the pre-emptive patch for 2.6.3 ? Lev 2004-04-15 10:20 pm Check your config in your favorite way. There’s a preemtive option there (however there are various known, unfixed glitches in it). 2004-04-18 12:32 am the only modern general purpose operating system in wide use using a microkernel are I guess Windows NT, and OSX. Heh. It also just so happens to be that that the only two operating systems in wide use happen to BE Windows NT, and OSX. >:} That was my evil grinch face, BTW.